Workload details

24h · 100%
1h12h24h
Commitment horizon
GB / mo
Pricing type
No tracked provider publishes reserved pricing for this GPU.

Estimated monthly cost

2× H200

24 hours a day is 730 GPU-hours a month, billed at on-demand.

  1. RunPod Community$5,241 / mo

    $3.59 /GPU-hr · global (uniform pricing)

    egress: not published

  2. Vast.ai$5,826 / mo

    $3.99 /GPU-hr · global (uniform pricing)

    egress: $3 (tier band)

  3. Crusoe$6,263 / mo

    $4.29 /GPU-hr · us-north-1 (uniform pricing)

    egress: not published

OpenMetalBare Metal$5,458 / mo

Fixed monthly, 1–2 × NVIDIA H200 NVL (PCIe), 141GB per card, egress included, 99.96%+ network SLA · Ashburn, VA / Los Angeles, CA

Commitment discounts are available. Request a quote to find the right pricing that suits you.

OpenMetal's H200 is the NVL PCIe card. Most rates above are for H200 SXM, a higher-bandwidth part; for NVLink-bound multi-GPU training the comparison is not like-for-like. Vast.ai's H200 NVL is the same part at $3.72/GPU-hr (marketplace, 1 offer).

Request a quote

Written quotes honoured for 30 days.

Compare all 9 H200 rates →

HOW TO USE THIS CALCULATOR

Estimate your GPU bill in three steps

1

Describe the workload

Pick the GPU, how many you need and how much data leaves the provider each month. Rates are normalised to one GPU-hour, so a count of one is a real number even where the smallest purchasable node holds eight.

2

Set utilisation and horizon

Hours per day is the input that decides everything. A commitment horizon selects the reserved rates that exist at that term, and the matching fixed-server term.

3

Read the breakeven

The estimate names the hours per day at which a fixed monthly server costs the same as renting. Below it, rent. Above it, owning the capacity is cheaper.

COST DRIVERS

What drives your GPU bill

Utilisation

The single largest lever. The same server is excellent value at 24 hours a day and poor value at four, because the fixed part of the bill does not shrink when you stop using it.

Pricing type

On-demand is the published rate. Spot is cheaper and interruptible, so it is not comparable to dedicated capacity. Reserved trades a commitment for a discount — and for several GPUs here, nobody publishes one at all.

Provider tier

Marketplace listings are individual hosts, priced by their owners and reclaimable at short notice. They are frequently the cheapest row on this page, and they are a different purchase from guaranteed capacity — worth choosing deliberately.

Egress

Hyperscalers meter data leaving their network per gigabyte. Some specialists publish no egress rate at all, in which case a total here is marked understated rather than quietly assumed to be free.

FREQUENTLY ASKED QUESTIONS

Frequently asked questions

How much does an NVIDIA H100 cost per hour?
Between $2.27 and $12.29 per GPU-hour across the 12 providers tracked here, median $4.74. That 5.42x spread is for the same card, so the provider matters more than almost anything else you choose.
How much does it cost to run a GPU for a month?
Which cloud provider has the cheapest GPUs?
Is it cheaper to rent GPUs by the hour or use a dedicated server?
How much does GPU data egress cost?
What does a price per GPU-hour mean?
Why is reserved pricing available for some GPUs but not others?
How current are these GPU prices?